SendMessageWrap

The SendMessageWrap class is an iterable wrapper for message lists sent to the model.

Description

SendMessageWrap (implementing Iterable[Message | ToolResult]) organizes the full message context into four parts: system train message, memory messages (without system message), the user_query, and end_messages. Iterating yields train, then memory, then user_query.

Properties

• train (Message[str]): System message
• memory (list[Message | ToolResult]): Messages without the system message
• user_query (Message): The user's query message
• end_messages (list[Message | ToolResult]): End messages (appended at the end)

Constructor

• __init__(train: dict | Message, memory: list | MemoryModel, user_query: Message | None = None): Builds the wrapper. If user_query is omitted, the last memory message is used; it must be a user message (raises ValueError otherwise), and is popped from memory when auto-detected

Methods

• classmethod validate_messages(messages: list) -> SendMessageWrap: Builds a wrapper from a message list, locating the system message as train (raises ValueError if none found)
• __len__() -> int: len(memory) + 2 + len(end_messages)
• __iter__(): Yields train, then memory, then user_query

Example

from amrita_core.types import SendMessageWrap

wrap = SendMessageWrap(
    train={"role": "system", "content": "You are a helpful assistant."},
    memory=[{"role": "user", "content": "Hello!"}],
)
for msg in wrap:
    print(msg.role, msg.get_content())